CyberLink shrinks facial recognition model by 98 percent for edge biometrics
Facial recognition and edge computing are the two most promising opportunities in the artificial intelligence market in 2019, CyberLink CEO Dr. Jau Huang told an audience at the COMPUTEX Forum 2019.
In his address on “Facial recognition at the Edge: The Future is Now,” Huang said that American facial recognition leaders are mostly focused on facial biometrics delivered from the cloud. Taiwan, meanwhile, where CyberLink is based, is a leader in the design and manufacture of IoT devices, which gives the company a natural niche in edge facial recognition, serving customers who prefer not to partner with Chinese providers, or use the cloud.
CyberLink’s algorithm was bested by entries from 17 vendors in the Visa category of the most recent NIST FRVT 1:1 report, and ranked 12th in the MegaFace Challenge in November, 2018, Huang says. The company’s FaceMe supports all major operating systems and platforms for integration in security systems, digital signage, kiosks, POS systems, and apps.
Huang discussed CyberLink’s efforts to optimize its biometric technology for edge devices, shrinking the model size from 250MB to only 4.6MB. This means edge devices can run CyberLink’s facial recognition, even without the most powerful processing available. He also presented several case studies showing AIoT developers applying the FaceMe engine to different scenarios.
“As they are quickly evolving and maturing, game-changing facial recognition and edge computing technologies are now among the most significant AI trends. When deployed on edge devices and without operating system limitations, facial recognition’s processing speed can be greatly enhanced, resolving latency problems caused by cloud computing,” said Dr. Huang. “CyberLink’s FaceMe has ranked among the top-tier facial recognition engines in a number of global competitions, including the National Institute of Standards and Technology Face Recognition Vendor Test (NIST FRVT 1:1) and the MegaFace Challenge, establishing our solution as one of the most accurate and fasted engines in the world.”
Despite officially launching FaceMe less than a year ago, CyberLink says it has coordinated with more than 100 global partners.
The company reported stronger-than-expected demand for FaceMe when announcing its fourth quarter 2018 earnings.
artificial intelligence | biometrics | biometrics at the edge | CyberLink | edge computing | facial recognition | IoT | Taiwan